AI Assist FAQs
FAQs for using AI Assist are identified in this topic.
What AI agents are available in Success Navigator through AI Assist?
Fusion: Feature Innovation, Roadmap, OMBP, Programs, Projects, Milestones, Production Profile, and User Profile agents are available.
OCI: OCI Knowledgebase agent is available.
What data sources are used to train the AI model?
The data sources used to train Oracle’s AI models are carefully selected to ensure security, privacy, and relevance. At a high level, Oracle’s models are trained on:
- Oracle proprietary data, such as Oracle product documentation on public-facing sites (docs.oracle.com), knowledge bases, Modern Best Practices, and Release Features from the Release Readiness site (https://docs.oracle.com/en/cloud/saas/readiness/index.html). This also includes Web Search from domains *.oracle.com. The OCI Knowledgebase agent is trained on MyLearn content.
- Customer data, used only with strict controls and permissions, including data within Success Navigator. Success Navigator does not connect to customer environments.
- Data sources within Success Navigator include: Journeys, Programs, Projects, Scope, Roadmaps, Release Features, User Profile, Production Profile, OMBPS including videos and transcripts, MyLearn video transcripts for OCI and Fusion, and OCA Taxonomy.
- Publicly available data, including industry standards and open datasets
The LLM models used are OCI Gen AI Service Pre-Trained models, utilizing the Chicago region for inference endpoint.
Pre-trained OCI GenAI models: https://docs.oracle.com/en-us/iaas/Content/generative-ai/pretrained-models.htm
Model availability per region: https://docs.oracle.com/en-us/iaas/Content/generative-ai/model-endpoint-regions.htm
Oracle does not train its enterprise AI models by scraping random external websites. All data sources are vetted, approved, and compliant with Oracle’s security and privacy policies. Models are regularly updated using new product documentation, release notes, and customer feedback, where permitted.
Describe the approaches used to detect the presence of personally identifiable information (PII) or sensitive data in the output of Oracle AI systems.
Oracle leverages automated tools and model-based techniques to identify and prevent the disclosure of PII or sensitive information in AI outputs. These include pattern-matching, natural language understanding, and post-processing filters to minimize the risk of exposing protected data.
How does AI Assist keep its knowledge base up to date and how often is it refreshed?
AI Assist’s knowledge base is regularly updated through a combination of bi-weekly data refreshes, integration with the latest Oracle documentation, and continuous monitoring of product updates and best practices. Oracle’s AI team ensures that the system reflects the most current features, security standards, and industry developments by syncing with official release notes, product documentation, and trusted sources. This helps ensure you always receive accurate and timely information when using AI Assist.
How accurate are AI Assist responses?
AI Assist responses are limited to content available only within the knowledge sources mentioned above. In the absence of content, AI Assist will respond to the user that there is not enough information available and refer to either Success Navigator or Oracle official sites for more information.
Would there ever be an instance where AI Assist guidance conflicts with official Oracle documentation?
AI Assist responses are limited to content available only within the knowledge sources mentioned above. However, it is possible that the knowledge sources could contain outdated content. We have updated the parameters included in the response to mitigate this where possible. There is also a disclaimer in the AI Assist window to advise the user to review for accuracy and refer to the original sources of the content. Large Language Models are generative which implicitly means that these are probabilistic models and cannot be relied on for a deterministic output in all scenarios.
How does Oracle detect or mitigate data poisoning or corrupted data?
Oracle employs rigorous data validation, monitoring, and filtering processes to detect anomalies or malicious patterns in training data. Regular audits and automated tools help identify and mitigate risks of data poisoning or corruption before they can impact the model.
How does Oracle ensure AI is used ethically and responsibly?
Oracle provides guidelines and best practices for ethical AI use, including transparency, fairness, and respect for privacy. Users are encouraged to follow these principles and to avoid using AI Assist for harmful or unethical purposes.
How does Oracle measure and validate the accuracy of the model, including monitoring for model/data drift, controlling for hallucinations, and addressing potential susceptibility to modifications based on queries by end users?
Model accuracy is measured using benchmark datasets, user feedback, and continuous evaluation. Oracle monitors for model and data drift and employs techniques to minimize hallucinations (incorrect or fabricated responses). The system is designed to prevent end-user queries from directly modifying the underlying model.
How does Oracle provide citations for the provided answers?
AI Assist provides in-line citations for facts, features, or claims, referencing official Oracle documentation or other trusted sources. Each citation is numbered and linked at the end of the response for easy verification.
How long does Oracle retain chat history?
You can delete your conversations at any point. Any conversations not deleted by the user are retained for a maximum of one year.
How secure is AI Assist?
AI Assist is built on Oracle Cloud Infrastructure, which provides robust security controls including encryption, access management, and continuous monitoring. All data is protected in transit and at rest, and the service complies with industry-leading security standards.
My company doesn’t allow use of AI. Can I turn off AI Assist?
Admins can enable or disable AI Assist. Open the profile menu, and then select the toggle next to AI Assist under My Organization.
What data is exposed, saved, etc?
When you use Success Navigator’s AI Assist features: Your queries and interactions may be logged for quality, troubleshooting, and improvement of the service, but these logs are handled according to Oracle’s privacy and security policies. Sensitive business data (like your transactions, HR records, etc.) is NOT shared with Oracle’s general AI training pipelines unless you have explicitly opted in. Data exposure is limited to what you have access to within your Oracle environment, and Oracle’s AI Assist does not ‘scrape’ or expose external or unauthorized data. Saved data: Some anonymized usage data may be retained for analytics or service improvement, but this is governed by Oracle’s data retention and privacy policies.
What languages can AI Assist translate to/from?
AI Assist can communicate in a variety of languages. Use the following prompt in AI Assist for a list of common languages that can be translated: “What languages can you translate?”.
Why am I seeing the message “Your terminology has been flagged as inappropriate. If this was in error, please rephrase your message and try again.”?
This message indicates that the language or terns you used may not align with Oracle's guidelines for appropriate and respectful communication, or may contain sensitive or restricted information. If you believe this is an error, please rephrased your message and try again. Such guardrails help ensure a safe and compliant environment for all users.